![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2640832%2F13f468c20cf5e2c0d8066c3b75d9e1ef%2FScreenshot%20from%202020-09-16%2011-45-15.png?generation=1600224590116972&alt=media\")
Good job this competition.
\nI was in medal line luckly because large shake up!
\nMy final best score is simple solution that I changed threshold of prediction of ResNeSt notebook.
\nPlease tell me what cause your shake up.
", "messages": [ { "id": "1012272", "postDate": "09/16/2020 02:21:53", "content": "Good job this competition.
\nI was in medal line luckly because large shake up!
\nMy final best score is simple solution that I changed threshold of prediction of ResNeSt notebook.
\nPlease tell me what cause your shake up.
", "rawMarkdown": "Good job this competition.\n\nI was in medal line luckly because large shake up!\n\nMy final best score is simple solution that I changed threshold of prediction of ResNeSt notebook.\n\nPlease tell me what cause your shake up.", "votes": null }, { "id": "1012309", "postDate": "09/16/2020 02:51:05", "content": "Hi @yoshito
\nI think it's probably due to the difference in the ratio of nocall between private and public LBs.
\nI tried to submit only nocall.
\nIn public, it was 0.544, but in private it was 0.577.
\nIn other words, the ratio of nocall is higher in private.
\nAs you know, threshold and nocall are closely related.
\nI think this is the reason.
\nthanks :)
If you do oof validation properly and don't overfitting to the LB. You will get a stable result.![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5600457%2Fd1f24cde1394d62b7b359b6274d7c31c%2FQQ20200916110512.png?generation=1600225539870520&alt=media\")
Thank you everyone!!
\nOnly nocall sumission is good information
\nI did not come up with good local validation…
\nWhich discussion is good for learning ???
We focused on finding mixed birds well with noise as our cv staggery. it helps us jump from 341 to 21 rank in private LB.
\n", "rawMarkdown": "We focused on finding **mixed birds** well with noise as our cv staggery. it helps us jump from 341 to 21 rank in private LB. \n- https://www.kaggle.com/c/birdsong-recognition/discussion/183209", "votes": null }, { "id": "1012747", "postDate": "09/16/2020 09:13:25", "content": "Thanks for sharing! This is interesting result
", "rawMarkdown": "Thanks for sharing! This is interesting result", "votes": null }, { "id": "1014317", "postDate": "09/17/2020 10:52:05", "content": "Thank you ! I will see later
", "rawMarkdown": "Thank you ! I will see later", "votes": null }, { "id": "1014319", "postDate": "09/17/2020 10:53:18", "content": "Yes, you are right,,,, but I did not come up with good validation idea.
\nCould you tell me your validation method?
Threshold was very important in this competition if you did not include a nocall model.
", "rawMarkdown": "Threshold was very important in this competition if you did not include a nocall model.", "votes": null } ], "comments": [ { "id": 1012309, "author_name": "fkubota", "author_url": "", "post_date": "09/16/2020 02:51:05", "content": "Hi @yoshito
\nI think it's probably due to the difference in the ratio of nocall between private and public LBs.
\nI tried to submit only nocall.
\nIn public, it was 0.544, but in private it was 0.577.
\nIn other words, the ratio of nocall is higher in private.
\nAs you know, threshold and nocall are closely related.
\nI think this is the reason.
\nthanks :)
Thanks for sharing! This is interesting result
", "votes": null, "replies": [] } ] }, { "id": 1012325, "author_name": "gzl0506", "author_url": "", "post_date": "09/16/2020 03:05:46", "content": "If you do oof validation properly and don't overfitting to the LB. You will get a stable result.
Yes, you are right,,,, but I did not come up with good validation idea.
\nCould you tell me your validation method?
Thank you everyone!!
\nOnly nocall sumission is good information
\nI did not come up with good local validation…
\nWhich discussion is good for learning ???
We focused on finding mixed birds well with noise as our cv staggery. it helps us jump from 341 to 21 rank in private LB.
\n", "votes": null, "replies": [] }, { "id": 1014317, "author_name": "yoshito", "author_url": "", "post_date": "09/17/2020 10:52:05", "content": "Thank you ! I will see later
", "votes": null, "replies": [] } ] }, { "id": 1016209, "author_name": "aliabdin1", "author_url": "", "post_date": "09/18/2020 18:30:46", "content": "Threshold was very important in this competition if you did not include a nocall model.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1012272": "Good job this competition.\n\nI was in medal line luckly because large shake up!\n\nMy final best score is simple solution that I changed threshold of prediction of ResNeSt notebook.\n\nPlease tell me what cause your shake up.", "1012309": "Hi @yoshito \n\nI think it's probably due to the difference in the ratio of nocall between private and public LBs.\n\nI tried to submit only nocall.\nIn public, it was 0.544, but in private it was 0.577. \nIn other words, the ratio of nocall is higher in private. \nAs you know, threshold and nocall are closely related.\nI think this is the reason.\nthanks :)\n\n", "1012325": "If you do oof validation properly and don't overfitting to the LB. You will get a stable result.", "1012501": "Thank you everyone!!\n\nOnly nocall sumission is good information\n\nI did not come up with good local validation...\nWhich discussion is good for learning ???", "1012706": "We focused on finding **mixed birds** well with noise as our cv staggery. it helps us jump from 341 to 21 rank in private LB. \n- https://www.kaggle.com/c/birdsong-recognition/discussion/183209", "1012747": "Thanks for sharing! This is interesting result", "1014317": "Thank you ! I will see later", "1014319": "Yes, you are right,,,, but I did not come up with good validation idea.\nCould you tell me your validation method?", "1016209": "Threshold was very important in this competition if you did not include a nocall model." }, "source": "meta" }